Help Center
5/6/2024
In a project involving concept mapping, participants are asked to provide individual responses to a given survey question. The responses are then cleaned and sorted to form the dataset that will be used for concept mapping analysis. The classic sorting method involves reading the responses manually and sorting them using a spreadsheet (e.g., Excel). This method is tedious and quickly becomes impractical with large numbers of participants. Another approach is to randomly select a number of responses. Although this option is quicker, it can be biased and may not reflect the diversity of the responses.
To improve this step, we propose using natural language processing methods to sort response items based on similarity. First, each sentence is converted into a numerical vector using a pre-trained language model, i.e., Sentence-BERT (Reimers and Gurevych, 2019). Next, the vectors are projected onto a reduced space using Uniform Manifold Approximation (Leland et al., 2020) and sorted through hierarchical clustering. This process significantly reduces the data cleaning and selection time by partitioning the response items according to their themes.
By using natural language processing methods, we can conduct concept mapping consultations with a larger number of participants and consider all responses when selecting items for analysis.
For more information on concept mapping and Polygon’s CM* tool, see the following links: What is concept mapping? and CM*
Concept Mapping Tool
Natural language processing
